small language model

Leveraging small language models for model predictive control of dynamic systems

Constitutional Midtraining: Content Presence Drives Alignment Gains

Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We build a 394M-token constitutional corpus from Anthropic's Constitution and apply constitutional midtraining at 120B scale, where principled, values-based content is inserted into midtraining. A 2x2 design (curriculum ordering x deliberative reasoning) was used to produce four constitutionally midtrained conditions, plus a control, which were evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment. All models were evaluated across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperformed the control on alignment generalization and durability, notably on blackmail: SFT instilled a blackmail propensity in all models, but constitutional midtraining blunted it, with the advantage surviving benign fine-tuning (-17.5pp). This durability did not extend to settings that required active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also mattered more than its structure, and constitutional midtraining incurred no capability cost, on average, at any stage (MMLU, ARC-Easy, piqa, GSM8K). A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.

Where A Small Language Model Helps in Invoice Categorisation, Understood Through Embedding Geometry

Analysis of the pre-trained embedding geometry of a small sentence transformer (SBERT) and classic SLM reveals that pre-trained embedding geometry is associated with classification performance and reveals a counterintuitive finding that a structured input that would help a human reader does not improve the SLM performance.

Policy-Invariant Reward Shaping from LLM Feedback: A Framework for Hybrid RL Agents

This work formalizes the hybrid LLM-planner and RL-controller architecture as a Goal-Augmented Markov Decision Process and shows that when the LLM per-state progress score is used as a bounded potential function, the resulting shaping term preserves the optimal policy set even when the LLM scores are inaccurate.

Understanding Multilingual Medical ASR Adaptation Through Layer-Wise Analysis

This paper investigates how multilingual medical adaptation reshapes the internal representations of Whisper models through layer-wise encoder analysis, and shows that English medical fine-tuning produces the dominant encoder shift, whereas multilingual continuation largely preserves the adapted representation space.

Beyond the Transcript: Detecting Covert Co ordination in Latent Multi-Agent Communication

Verifiable Latent Alignments (VLA), an activation-aware framework for monitoring and steering these private communication channels, is introduced and shows that the evaluated private channel attacks can be monitored without training the primary monitor on attack examples and mitigated when matched counterfactual access is available.

Breaking the weakest link to evade vision language models

To efficiently generate adversarial examples, a gradient-based attack method is proposed that performs optimization exclusively on the vision encoder of the VLM rather than on the entire multimodal architecture, which significantly reduces the computational cost and resource requirements of the attack while maintaining strong effectiveness.

TestifAI: Tomography-Based Testing for Deep Learning Systems

TestifAI, a deep learning testing framework for efficient and accurate estimation of robustness against combinations of perturbations, is proposed and partial model tomography is introduced, a novel approach to reconstructing model behaviour in a multi-perturbation space from tests that apply only a small number of perturbations.

Improving Natural-Language Combinatorial-Optimization Accuracy in Resource-Constrained Language Models via Formal Abstractions

SDDL is introduced, a neuro-symbolic framework that translates natural-language scheduling problems into compact, solver-aligned representations of tasks, resources, constraints, and objectives, while delegating low-level modeling and search to a deterministic compiler and external solver.

LLMs Leak Training Data Beyond Verbatim Memorization: Extraction via Membership Decoding

The Membership Decoding method is a plug-and-play replacement for standard decoding that requires only black-box token probabilities, and a new token-level membership inference method is proposed by leveraging likelihood from reference models, shifting the generation from the original token distribution to the member token distribution.